fix(search): limit same_entity recency marker to diff/doc rows

src/aggregate.test.ts

@@ -245,28 +245,46 @@ describe("duplicate rate regression (issue #189 measurements, 2026-08-01)", () =
 });
 
 describe("markRecency (issue #267)", () => {
+  it("marks nothing for an issue and its comments (issue #269)", () => {
+    const m = markRecency([
+      { id: "issue", type: "issue", updatedAt: "2026-09-15T00:00:00Z" },
+      { id: "c1", type: "issue_comment", updatedAt: "2026-09-20T00:00:00Z" },
+      { id: "c2", type: "issue_comment", updatedAt: "2026-09-29T00:00:00Z" },
+    ]);
+    expect(m.size).toBe(0);
+  });
+  it("compares only diff/doc rows in a mixed group (issue #269)", () => {
+    const m = markRecency([
+      { id: "doc", type: "doc", updatedAt: "2026-09-15T00:00:00Z" },
+      { id: "diff", type: "diff", updatedAt: "2026-09-20T00:00:00Z" },
+      { id: "other", type: "issue", updatedAt: "2026-09-29T00:00:00Z" },
+    ]);
+    expect(m.get("doc")).toBe("superseded");
+    expect(m.get("diff")).toBe("latest");
+    expect(m.has("other")).toBe(false);
+  });
   it("marks older rows superseded and newest latest", () => {
     const m = markRecency([
-      { id: "old", updatedAt: "2026-09-15T00:00:00Z" },
-      { id: "new", updatedAt: "2026-09-29T00:00:00Z" },
-      { id: "mid", updatedAt: "2026-09-23T00:00:00Z" },
+      { id: "old", type: "diff", updatedAt: "2026-09-15T00:00:00Z" },
+      { id: "new", type: "diff", updatedAt: "2026-09-29T00:00:00Z" },
+      { id: "mid", type: "diff", updatedAt: "2026-09-23T00:00:00Z" },
     ]);
     expect(m.get("old")).toBe("superseded");
     expect(m.get("mid")).toBe("superseded");
     expect(m.get("new")).toBe("latest");
   });
   it("marks nothing for a single row or equal timestamps", () => {
-    expect(markRecency([{ id: "a", updatedAt: "2026-09-15T00:00:00Z" }]).size).toBe(0);
+    expect(markRecency([{ id: "a", type: "diff", updatedAt: "2026-09-15T00:00:00Z" }]).size).toBe(0);
     expect(markRecency([
-      { id: "a", updatedAt: "2026-09-15T00:00:00Z" },
-      { id: "b", updatedAt: "2026-09-15T00:00:00Z" },
+      { id: "a", type: "diff", updatedAt: "2026-09-15T00:00:00Z" },
+      { id: "b", type: "diff", updatedAt: "2026-09-15T00:00:00Z" },
     ]).size).toBe(0);
   });
   it("leaves unparsable timestamps unmarked", () => {
     const m = markRecency([
-      { id: "a", updatedAt: "" },
-      { id: "b", updatedAt: "2026-09-15T00:00:00Z" },
-      { id: "c", updatedAt: "2026-09-16T00:00:00Z" },
+      { id: "a", type: "diff", updatedAt: "" },
+      { id: "b", type: "diff", updatedAt: "2026-09-15T00:00:00Z" },
+      { id: "c", type: "diff", updatedAt: "2026-09-16T00:00:00Z" },
     ]);
     expect(m.has("a")).toBe(false);
     expect(m.get("c")).toBe("latest");

src/aggregate.ts

@@ -137,10 +137,14 @@ export type Recency = "latest" | "superseded";
  * "latest" for the newest. Returns an empty map when the group holds fewer
  * than two distinct timestamps — the pool proves nothing about recency then.
  * Rows with an unparsable timestamp are left unmarked.
+ * Only `diff` and `doc` rows take part (issue #269): an issue's comments are
+ * separate events, not versions of the issue, so other types are never
+ * marked and never enter the comparison.
  */
-export function markRecency(rows: Array<{ id: string; updatedAt: string | null | undefined }>): Map<string, Recency> {
+export function markRecency(rows: Array<{ id: string; type: string; updatedAt: string | null | undefined }>): Map<string, Recency> {
   const out = new Map<string, Recency>();
   const times = rows
+    .filter((r) => r.type === "diff" || r.type === "doc")
     .map((r) => ({ id: r.id, t: r.updatedAt ? Date.parse(r.updatedAt) : NaN }))
     .filter((r) => !Number.isNaN(r.t));
   if (new Set(times.map((r) => r.t)).size < 2) return out;

src/mcp.ts

@@ -338,8 +338,8 @@ export function createRagMcpServer(env: Env): McpServer {
         "Results are aggregated per underlying entity: a file's doc row and its commit diffs are one result, " +
         "an issue or PR and its comments / reviews are one result. top_k therefore counts distinct entities, " +
         "and a result that absorbed others carries same_entity { count, others[] } with links to them. " +
-        "When those rows hold more than one distinct updated_at, the representative and each others entry carry recency: " +
-        "\"superseded\" = a newer row of the same entity is in the pool, \"latest\" = the newest row in the pool. " +
+        "Among diff and doc rows only, when those rows hold more than one distinct updated_at, each such row (representative or others entry) carries recency: " +
+        "\"superseded\" = a newer diff/doc row of the same entity is in the pool, \"latest\" = the newest such row in the pool. Issues, PRs, comments and reviews never carry recency. " +
         "Absent recency is not a claim of being latest, and recency never affects ranking or scores.\n" +
         "Every result row — and every same_entity.others entry — carries vector_id, the handle mode 4 takes. " +
         "It is a handle for reaching a row you just found, not a durable identifier: the id scheme has been " +
@@ -943,8 +943,11 @@ export function createRagMcpServer(env: Env): McpServer {
         // dropped) so the caller can still reach every version / comment.
         const folded = collapsedInto.get(f.vectorId) ?? [];
         const recency = markRecency([
-          { id: f.vectorId, updatedAt: r.updatedAt },
-          ...folded.map((o) => ({ id: o.vectorId, updatedAt: resolveRow(payload.get(o.vectorId)).updatedAt })),
+          { id: f.vectorId, type: r.type, updatedAt: r.updatedAt },
+          ...folded.map((o) => {
+            const or = resolveRow(payload.get(o.vectorId));
+            return { id: o.vectorId, type: or.type, updatedAt: or.updatedAt };
+          }),
         ]);
         const sameEntity =
           folded.length > 0